Non classé

Why AI Projects Should Be Built Like Supply Chains

Published

7 heures ago

14 juillet 2026

An enterprise team begins an artificial intelligence project with a familiar question: Which model should we use? The team compares benchmarks, studies pricing, debates whether the newest large model is worth the additional cost, and eventually selects a platform.

The first outputs arrive, and some are impressive. Others contain unsupported claims, repeat information from earlier sections, ignore important instructions, or vary sharply in quality from one run to the next.

The team responds by expanding the prompt and adding more context. Before long, the model is being asked to research the subject, verify facts, organize findings, write the report, match the company’s voice, improve readability, optimize the structure, check for duplication, and ensure that nothing important is missing.

The prompt grows longer, but the underlying problem remains. The project has been designed around a single model interaction even though the work itself is not a single task.

It is a process, and supply chain professionals should recognize the mistake immediately. No modern manufacturer expects one machine to receive raw materials, fabricate every component, assemble the finished product, inspect it, package it, and prepare it for delivery.

Complex production is divided into specialized stages because each stage requires different capabilities, controls, and measures of quality. Enterprise AI is moving toward the same realization.

The future of AI will not be defined only by larger models. It will be defined by better production systems.

From a Prompt to a Production Line

Imagine that the enterprise team starts again, but this time it does not ask one model to produce the finished report from a blank prompt. Instead, the work is divided into a sequence of specialized stages.

The first stage collects source material. The second extracts relevant facts and records where they came from.

The third stage classifies those facts and identifies relationships between them. The fourth builds a structured outline that reflects the intended audience, argument, and format.

A drafting model then turns that outline into prose. A stronger editorial model reviews the completed draft for clarity, consistency, factual alignment, and tone.

Finally, deterministic software checks measurable requirements such as word count, metadata, links, formatting, duplicate language, and required sections. The result is not simply another prompt but an AI production line.

Each stage has a defined input and output, and each handoff can be inspected before the work continues downstream. That changes the nature of the system.

When an output is weak, the team no longer has to guess what went wrong. It can determine whether the problem originated in research, classification, outlining, drafting, editing, or quality assurance.

A missing fact can be traced back to retrieval, while a weak structure can be corrected in the outline. Repetitive writing can be addressed during editorial review, and a broken link can be fixed during quality assurance without regenerating the entire document.

The system becomes diagnosable, which is one of the defining differences between an experiment and an industrial process. Once defects can be traced to specific stages, they can be corrected systematically rather than treated as random model behavior.

The Lesson Supply Chains Already Learned

Supply chains became more capable by abandoning the idea that one facility, one supplier, or one process should do everything. Specialization improved performance, while standardization improved handoffs.

Quality gates prevented defects from moving downstream, and visibility made it possible to identify bottlenecks. Redundancy reduced dependence on a single point of failure, while continuous improvement raised performance over time.

The same principles increasingly apply to AI. A low-cost model may be well suited to extracting structured facts from a set of documents, while a more capable model may be better at synthesis and editorial judgment.

Python or another deterministic tool may be more reliable for validation, deduplication, calculations, and file handling. A knowledge graph may be better than a language model at preserving relationships between entities.

A human expert may remain essential when the decision involves ambiguity, risk, or strategic judgment. The objective is not to force one model to perform every task but to orchestrate the best combination of models, software, data, and human expertise.

That is fundamentally a supply chain problem. It involves routing work through the right sequence of specialized resources to produce a reliable outcome.

Cost Savings Are Only the Beginning

The immediate financial argument for this approach is easy to understand. Organizations do not need to use their most expensive model for every step.

Routine extraction, classification, and formatting can often be handled by smaller models or deterministic software. More capable models can then be reserved for the moments where reasoning, synthesis, or editorial judgment create the greatest value.

That can substantially reduce token costs, but cost reduction is not the most important benefit. The larger advantage is control.

A one-step system asks a model to interpret a broad objective and quietly make hundreds of intermediate decisions. Those decisions are usually invisible to the user.

When the final answer is wrong, inconsistent, or incomplete, there may be no clear way to determine why. A multi-step system exposes those decisions and makes them easier to inspect.

Research can be reviewed before drafting begins, classifications can be tested against a taxonomy, and claims can be linked to sources. Drafts can be compared with the underlying evidence, while quality rules can be applied consistently across every output.

The organization is no longer merely generating content. It is managing a production process.

The Value of the Work in Progress

In traditional manufacturing, work in progress is usually viewed as inventory that must be controlled. In AI production, the intermediate work can become an asset of its own.

Consider a company building a supplier directory. A one-step workflow might ask a model to visit a supplier’s website and write a profile.

When the profile is complete, the underlying research effectively disappears inside the finished prose. The final page may be useful, but the evidence and structure used to create it are difficult to reuse.

A multi-step workflow would first create a structured supplier packet containing the company’s canonical name, capabilities, products, industries served, geographic presence, source links, category assignments, confidence levels, and unresolved questions. That packet can support the supplier profile, but it can also support much more.

It can populate a comparison table, connect the supplier to an industry ontology, support a market map, feed a research report, improve internal linking, and provide context to an AI assistant. It can also be updated later without repeating the entire research process.

The article or profile becomes one expression of the knowledge rather than the knowledge itself. This is an important distinction because the most valuable output of an AI system may not be the document that appears on the screen.

The more durable asset may be the structured knowledge created along the way. That knowledge can support multiple products, channels, and future workflows.

Quality Cannot Be Added at the End

Many enterprises still treat quality control as a final review step. The model produces an answer, and a human is asked to check it.

That approach does not scale well. A reviewer examining one report can catch obvious problems, but a reviewer responsible for hundreds or thousands of pages cannot reconstruct every source, validate every classification, compare every phrasing pattern, and confirm every piece of metadata.

Quality must therefore be designed into the process. That means validating source quality before information enters the system and distinguishing verified facts from model inference.

It also means using structured schemas so required fields cannot quietly disappear. Confidence thresholds should be applied, and uncertain cases should be escalated for human review instead of being treated as equally reliable.

Not every stage should be handled by generative AI. Language models are strong at interpretation, synthesis, and expression, but they are less dependable when exact counting, deterministic comparison, or strict rule enforcement is required.

A mature architecture uses generative AI where judgment is needed and conventional software where certainty is required. The same principle applies in supply chains, where inspection cannot compensate for a production system that repeatedly introduces defects.

Quality must be built upstream. The earlier a defect is found, the less damage it can cause downstream.

Visibility Changes Management

For decades, supply chain leaders have invested in visibility because they understand that an organization cannot manage what it cannot see. The same is true for AI workflows.

A black-box interaction provides very little operational visibility. A request goes in, an answer comes out, and the intermediate work remains hidden.

A staged system can reveal which sources were retrieved, which facts were accepted, and which claims had low confidence. It can also show which model handled each stage, where the final reviewer made changes, and which outputs repeatedly failed quality checks.

This creates the possibility of managing AI performance rather than merely observing it. Over time, the organization can identify recurring defects, improve prompts, replace weak models, update templates, refine taxonomies, and strengthen source selection.

The AI system begins to improve not only through model upgrades but through process improvement. That distinction is important because an enterprise can strengthen the overall system even when the underlying models remain unchanged.

Resilience Matters in AI Too

Dependence on one model also creates a form of concentration risk. A model may become more expensive, its behavior may change, its context limits may become restrictive, or its availability may decline.

A new model may outperform it in one task but not another. Organizations that build their entire workflow around one provider or one model may therefore find it difficult to adapt.

A modular pipeline is more resilient. The extraction model can be replaced without redesigning the editorial stage, and the review model can be upgraded without rebuilding the research process.

A vector database can be changed while preserving the output schema. A human approval step can also be added to a high-risk workflow without disrupting the rest of the architecture.

This modularity resembles a well-designed supply network in which components can change while the broader system continues to operate. The objective is not merely efficiency but adaptability.

The Model Is Not the System

The first phase of enterprise AI encouraged organizations to think of the model as the product. That was understandable because the model was the most visible and impressive component.

A model alone, however, is no more a complete enterprise system than an engine is a complete transportation network. The value comes from what surrounds it.

That includes the data entering the system, the context provided to the model, the tools it can call, and the rules governing its behavior. It also includes the mechanisms that validate its output, the people who supervise important decisions, the knowledge retained after the task is complete, and the feedback used to improve future performance.

This is why the question “Which model should we use?” is becoming less useful on its own. A better question is, “How should the work move through the system?”

That question forces the enterprise to think about architecture, handoffs, quality, governance, and continuous improvement. It shifts the conversation from model selection to operating design.

Building Intelligence as a Supply Chain

The analogy is not rhetorical because it offers a practical blueprint. Raw information enters the system in much the same way raw material enters a production network.

Retrieval and extraction prepare it for use, while classification and ontology assign meaning and structure. Planning organizes the work into a viable production sequence.

Generation transforms the structured inputs into a usable product, while editorial review improves the finish. Quality assurance checks conformity, and human experts manage exceptions.

Performance data then flows back upstream to improve the next production cycle. Seen this way, AI is not simply answering questions.

It is converting information into decisions, documents, recommendations, and knowledge assets through a coordinated sequence of transformations. That is precisely what supply chains do with physical goods.

The Next Competitive Advantage

Access to powerful models will continue to broaden, and the models themselves will continue to improve. Prices will change, benchmark leaders will rotate, and new providers will emerge.

As that happens, access to a particular model will become less of a durable advantage. The more defensible capability will be the system built around the models.

Organizations that develop reliable research packets, proprietary taxonomies, structured knowledge bases, quality-control rules, reusable workflows, and feedback loops will be able to produce stronger results regardless of which model happens to lead the market at a given moment. Their advantage will come from orchestration rather than access.

They will know how to route work, which tasks require premium reasoning, and which do not. They will also know where human judgment creates value and how to preserve knowledge instead of discarding it after every interaction.

They will improve the process with every production cycle. That is the larger lesson.

The future of enterprise AI will not belong to the companies that simply buy access to the largest model. It will belong to the companies that build the best intelligence supply chains.

Those companies will treat information as an input, knowledge as an asset, quality as a process, and AI as a coordinated production system rather than a single prompt. Supply chain leaders have spent decades learning how to design systems that are specialized, visible, resilient, and continuously improving.

Those same principles may now become some of the most important principles in enterprise AI.

The post Why AI Projects Should Be Built Like Supply Chains appeared first on Logistics Viewpoints.

Non classé

Container rates starting to spike on peak season rush – June 2, 2026 Update

Published

4 heures ago

14 juillet 2026

Weekly highlights

Ocean rates – Freightos Baltic Index

Asia-US West Coast prices (FBX01 Weekly) increased 1%.

Asia-US East Coast prices (FBX03 Weekly) increased 4%.

Asia-N. Europe prices (FBX11 Weekly) increased 3%.

Asia-Mediterranean prices(FBX13 Weekly) increased 1%.

Air rates – Freightos Air Index

China – N. America weekly prices increased 1%.

China – N. Europe weekly prices decreased 6%.

N. Europe – N. America weekly prices decreased 2%.

Analysis

Approaching 100 days since the start of the Iran war, despite periodic reports that an agreement that would open the Strait of Hormuz is near, the sides continue to exchange fire and sanctions, and the waterway remains closed.

For the container market, the closure has primarily meant upward pressure on freight rates via carriers passing on war-elevated fuel costs, which manifested in different ways on different lanes during the low demand months of March, April and most of May this year.

Join 70,000+ Supply Chain Experts Who Never Miss an Issue!

Start your week with the industry insights others miss.

« * » indicates required fields

This field is for validation purposes and should be left unchanged.

Email*

Consent*

I agree to receive logistics, freight tech, and commercial emails.

But peak season demand is kicking in early on east-west lanes, with reports of contracted shippers already seeing allocations reduced and premiums applied. So spot rates that climbed moderately – about 15% – across the ex-Asia lanes through mid-May GRIs to levels around 20% higher than a year ago, are starting to spike this week.

Weekly averages for last week were about level to close out the month, with transpacific rates at about $3,200/FEU to the West Coast and $5,000/FEU to the East Coast, and Asia – Europe prices at about $3,000/FEU to N. Europe and $4,400/FEU to the Mediterranean. But June 1st GRIs and PSS introductions have daily rates spiking from $1,000/FEU to $1,800/FEU so far this week on these trades, with additional significant increases announced for mid-month across these lanes as well.

Daily rates for Asia – Europe lanes have already surpassed peak season highs from last June/July, with transpacific still about $1,000/FEU short of last year’s brief, tariff frontloading-driven rate spike in July. Pre-existing war-related congestion in some tranship hubs, as well as rail congestion in Germany could also be a factor for rate pressure or delays for the relevant trades.

In trade war developments, IEEPA refunds – totalling about half of the total $166B paid – are on the way for importers whose customs entries had not already been liquidated, or finalized, by US Customs and Border Protection. But the Trump Administration indicated last week that it may challenge refunds for liquidated entries, arguing that the CBP is unauthorized to reliquidate and refund closed out entries without importer-specific court orders instructing it to do so.

Check out our full IEEPA tariff refund explainer and update page here.

This challenge, if successful, could mean that these importers would need to sue the government in trade court in order to get these duties refunded, and even if unsuccessful could mean a longer wait for impacted importers while the legal issues get sorted out. In the meantime, some trade law experts are advising importers with liquidated entries to file protests if the window hasn’t closed yet.

The trade war has resulted in lower or flat import volumes to the US alongside trade diversions driving volume increases between other countries as global players seek closer ties and trade growth beyond the US. Asia – Europe trade for example grew significantly last year and continues on pace so far in 2026. Even so, trade tensions between China and the EU may be increasing, as the EU considers legislation to curb subsidized imports.

Part of this issue relates to e-commerce imports to EU countries, which continue to grow significantly even as they flatten to the US and are reflected in diverging freighter capacity trends on these lanes. The EU will introduce a flat 3 EUR fee for low value imports starting in July, and a 2 EUR handling fee in November.

Though not as extensive as the US de minimis cancellation, these moves are likely to reduce EU e-commerce volumes arriving by air to some extent. Parcel carriers are warning that the system is still not ready for the new reporting requirements that will accompany the fee introductions, and warn of delays at European borders if these take effect in July.

Air cargo rates were about level on most major lanes this week, though the Freightos Air Index global benchmark – which is about even with April levels – remains more than 30% higher than before the start of the Iran war and year on year as capacity reductions and elevated jet fuel prices continue to impact price levels.

The post Container rates starting to spike on peak season rush – June 2, 2026 Update appeared first on Freightos.

Non classé

Ocean Freight Rates & Shipping Guide

Published

5 heures ago

14 juillet 2026

Latest Ocean Freight Rate News

Transpacific ocean freight rates have been falling since Lunar New Year, with Asia-US West Coast prices down 7% and East Coast down 5% last week according to Freightos Baltic Index data. This despite higher shipping volumes than last year due to tariff frontloading. The approaching April 2nd tariff announcement deadline could significantly impact shipping rates and patterns.

Ocean/Sea Freight Shipping Rates

When you start to ship freight at high volumes, it’s time to consider ocean freight. Here is your guide to everything ocean, from choosing the mode that’s right for you to calculating costs and transit times.

How much will your shipment cost? You can use this free calculator to get instant ocean freight estimates.

Loading…

What are Freight Shipping Rates?

Freight shipping rates are the costs of transporting cargo using ocean, air, rail, or road. These rates can vary significantly depending on mode of transport, distance, shipment volume, weight, and dimensions, as well as market conditions and seasonal fluctuations.

When it comes to ocean freight rates, several key components make up the total cost:

Base freight rate: The basic cost of shipping your goods from the port of origin to the port of destination.
Bunker Adjustment Factor (BAF): A surcharge that accounts for fluctuations in fuel prices.
Currency Adjustment Factor (CAF): A surcharge that compensates for exchange rate fluctuations.
Terminal Handling Charges (THC): Fees charged by the port authorities for handling containers at the origin and destination ports.
Surcharges: Various additional fees that may apply, such as for hazardous materials, peak season, or congestion at ports.

Working with experienced freight forwarders can help you navigate the complexities of freight rates and find the most cost-effective solution for your shipment. Platforms like Freightos.com allow you to compare rates from multiple providers instantly, making it easier to make informed decisions and optimize your shipping costs.

Looking for ocean freight rates?

Compare ocean rates from dozens of vetted providers

Compare Ocean Rates

Freightos – The Digital Freight Shipping Platform: Costs, Prices, Rates, and More.

Instantly compare ocean freight shipping rates with freight quotes from vetted providers. Find the balance of price and transit time that works for your ocean freight.

Our Ocean Freight Shipping Service

Freightos.com offers a comprehensive range of ocean freight shipping services, including instant quotes, freight forwarder comparison, online booking, customs clearance, cargo insurance, and shipment tracking.

As a global freight marketplace, we allow importers and exporters to choose from a variety of freight shipping options based on their specific needs. Freightos.com’s user-friendly interface and advanced technology also make it easy for small and large businesses to manage their freight shipments efficiently and cost-effectively. Discover how our reliable and seamless freight shipping service can simplify your logistics, providing the support you need for smooth operations.

LCL Shipping

Freightos.com offers a range of LCL (less-than-container load) shipping services to businesses looking to ship smaller quantities of cargo.

We provide instant quotes for LCL shipments, allowing businesses to compare rates from multiple forwarders and choose the best option based on their needs. Additionally, Freightos.com allows customs booking in-platform and easy communication with freight forwarders to help ensure that importers and exporters comply with all necessary regulations and requirements for LCL shipments.

FCL Shipping

For importers and exporters who need to transport larger quantities of cargo, Freightos.com offers a range of FCL (full container load) shipping services that include instant quotes for a variety of container types and sizes. Freightos.com can assist businesses with FCL shipping needs by providing instant quotes, a variety of container types and sizes, and support for customs clearance and documentation.

Ocean Freight Forwarders

Freightos.com works with many of the top and best ocean freight forwarders in the world.

The platform partners with leading freight forwarders to provide businesses with a wide range of shipping options, for both door-to-door and port-to-port shipments. Freightos.com’s advanced technology and online platform make it easy for businesses to compare rates and book freight shipments with its network of vetted forwarders. Our team of experts work closely with our forwarder partners to ensure that importers and exporters receive the highest quality of service throughout the shipping process.

Container Rates on Popular Routes

This data is based on Freightos Terminal.

To protect the underlying data, results here may vary slightly from the actual data points.

What is Ocean Freight?

Ocean freight transport is the shipping of goods by sea via shipping containers.

Ocean freight is the most common mode of transport that importers and exporters use. In fact, a full 90% of goods are shipped by ocean freight and sea freight. The other international freight transport modes (courier, air freight, express) are all faster, but they are also more expensive. Smaller shipments, and products with a high value, generally go by these other modes.

How Does Ocean Freight Work?

When you choose to ship your goods with ocean freight, your products will be packaged and possibly palletized either at the factory or by a third party. Your freight forwarder books space on a container vessel and your goods are shipped to the port to undergo a customs exam at the point of origin. Goods are then containerized into full containers or shared containers depending on whether you are shipping FCL or LC. Then the cargo is loaded onto ship for transportation.

Once the ship arrives at the destination port, goods pass through customs and once any duties and taxes are paid, are released. At this point, your goods will be shipped to a warehouse to be delivered to the final customer.

What Does Ocean Freight Mean?

Ocean freight means transporting goods through designated sea lanes by container vessel. This link in the supply chain is vital to cross-border trade that facilitates the movement of massive amounts of goods between countries.

There several shipping options available depending on the type of goods you are shipping. Full container load (FCL) shipping is when goods are containerized and shipped using standard sized 20 or 40 ft containers. For smaller quantities, LCL – or less than container load – means that shippers share container space since their volumes aren’t sufficient to fill a full container independently.

Ocean freight isn’t the only way to transport goods: for small, light, or high value products, many importers choose to ship by air. Air cargo is more expensive, but is faster and more secure. It’s also important to know that regulations for air cargo are more stringent than for ocean freight.

Freight Shipping by Sea

Capacity and Value – One container can hold 10,000 beer bottles! And ocean freight is cheaper. As a rule of thumb, any shipment weighing more than 500 kg is too expensive for air freight. For light shipments, use this chargeable weight calculator to work out whether your freight shipment will be charged by actual weight or dimensional weight. For live international shipping rates see our FBX index.
Fewer restrictions – International law, national law, carrier organization regulations, and individual carrier regulations all play their part in defining and restricting what goods are considered dangerous for transport. Generally, more products are restricted as air cargo than as ocean freight, including gases (e.g. lamp bulbs), all things flammable (e.g. perfume, Samsung Galaxy Note 7), toxic or corrosive items (e.g. batteries), magnetic substances (e.g. speakers), oxidizers and biochemical products (e.g. chemical medicines), and public health risks (e.g. untanned hides). For further information check out the Hazardous Material Table.
Emissions – CO2 freight emissions from ocean freight is minuscule compared with air freight. For example, according to this research, 2 tonnes shipped for 5,000 kilometers by ocean freight will lead to 150 kg of CO2 emissions, compared to 6,605 kg of CO2 emissions by air freight shipping.

What are the downsides of Ocean Freight?

Speed – Airplanes are about 30 times faster than ocean liners; passenger jets cruise at 575 mph, while slow-steaming ocean liners move at 16-18 mph. No surprise then, that a shipment going by air freight from China to the US usually takes at least 20 days more than by ocean freight.
Reliability – Port congestion, customs delays, and bad weather conditions generally add much more days to ocean freight than air freight. To date, tracking technology in air freight is often more advanced than ocean freight. That means that ocean freight is more likely to get misplaced than air freight. This is especially true when the ocean shipment is less than a container load. That said, ocean freight is becoming more reliable thanks to digitization.
Protection – Ocean freight is more likely to get damaged or destroyed than air cargo. That’s because it is in transit a lot longer, and because ships are more subject to movement. But don’t worry too much about ocean cargo falling off ships. The urban myth says 10,000 lost per year, but it’s more like 546 of the 120 million container movements per year that fall in the drink. Even less likely is piracy. Hotspots in recent years have included the Horn of Africa, the Gulf of Guinea, and the Malacca Straits.

Ocean Freight Services

Ocean and sea freight services break down to two further options: a full container load (FCL) and a less than container load (LCL). With LCL, several shipments are packed into one container. This means more work for the forwarder, there’s extra paperwork involved, as well as the physical work of consolidating various shipments into a container before the main transit and de-consolidating the shipments at the other end. This gives LCL three disadvantages:

LCL takes more time to deliver than an FCL shipment. It’s typically recommended to allow an extra one or two weeks for LCL.
There is an increased risk of damage, misplacement, and loss with LCL.
LCL costs more per cubic meter.

Since shipping rates are lower for FCL, it may be worth using a full container once your freight shipment is large enough, even if your goods do not fill a full container. The tipping point for upgrading from LCL to FCL (the smallest sized container is a 20 footer) is somewhere around 15 cubic meters.

Sea Freight Rates Per KG

With the exception of particularly heavy goods, most LCL is priced per volume of goods, and not by weight.

For most products, use these rules of thumb for which selecting the most cost-effective mode:

Freight shipments weighing more than 500 kg becomes uneconomic to go by air freight.
Ocean freight is around $2-$4/kg, and a China-US shipment will take around 30-40 days or more.
At about $5-8 per kilo, a China-US shipment between 150 kg and 500 kg can economically go air freight and will take around 8-10 days.
Express air freight is a few days quicker, but more expensive.
Packages that are lighter than 150 kg can economically go by courier (express freight).

Common Ocean and Sea Freight Costs, Rates, and Charges in Your Freight Quote:

Expect to see these items on ocean freight quotes and invoices:

Customs security surcharges (AMS, ISF)
Container Freight Station (these are the consolidation charges, and apply for LCL only)
Terminal Handling charges (charges by the port authority)
Customs brokerage
Pickup and delivery
Insurance
Accessorial charges (fuel surcharges, handling hazardous materials, storage, etc)
Routing charges (e.g. Panama Canal, Alameda Corridor)

Ocean Freight FAQs

Why do ocean freight quotes for the same shipment vary so much between providers?

Ocean freight quotes often vary because of differences in service levels and because quotes are not always directly comparable.

Not all freight forwarders have the same ability to secure space with carriers or offer the same level of support. Higher quotes may reflect stronger booking power, more reliable capacity, or additional services, while lower quotes may come with fewer included services or less support.

Just as often, quotes aren’t apples to apples. One may be door-to-door while another is port-to-port, assume a different Incoterm, or include services like inland transport or handling that others do not. Market conditions also play a role, as available space and seasonal demand can change what forwarders are able to quote at any given time.

Because of this, comparing quotes by email can be frustrating. Marketplaces like Freightos help by standardizing what’s being quoted upfront, making it easier to compare prices based on the same service scope.

How can I tell if my ocean freight quote is reasonable for my route and season?

The best way to judge whether a quote is reasonable is to compare multiple quotes rather than relying on a single price. Looking at several offers helps you understand the current market range for your route and timing.

If a quote is much higher than the rest, that can be a red flag – but prices that seem unusually low can also be risky, as they may come with limited service or additional fees added later. What matters most is where a quote sits relative to others for the same shipment details.

Using a marketplace like Freightos makes this comparison easier by showing multiple quotes at once for the same service scope, so you can quickly see where the market is. For businesses that want deeper insight into seasonal trends or route-specific shifts, tools like Freightos Terminal provide historical and real-time market data to help put individual quotes in context.

What is included (and not included) in a door-to-door ocean freight rate?

A door-to-door ocean freight rate typically includes the main transportation legs needed to move cargo from origin to destination. This often covers inland transport to the origin port, export handling and port fees, the ocean freight itself, port handling at the destination, and final delivery to the consignee’s location.

What’s not always included are GRIs, or costs that depend on the shipment, destination, or regulatory requirements. Customs duties and tariffs are usually paid separately, as are cargo insurance and optional services. Additional fees can also apply if special services are needed, such as liftgate delivery, appointments, or non-standard handling, and these are often only included if they’re requested upfront.

Because inclusions can vary by provider, it’s important to confirm exactly what’s covered in a “door-to-door” quote before booking.

Should I choose FCL or LCL for my shipment?

The choice between FCL (full container load) and LCL (less than container load) usually comes down to shipment size, timing, and reliability needs.

FCL is generally the better option if your shipment is large enough to justify a full container, or if reliability and predictability are especially important. Because the container is dedicated to a single shipper, FCL can be easier to plan around and may offer more consistent transit and handling, particularly during periods of congestion or tight capacity.

LCL is often a better fit for smaller shipments, whether that’s because you’re a smaller importer, you ship in smaller or more frequent batches, or your business is highly seasonal. Some shippers also use LCL strategically to split shipments or reduce exposure to market volatility, even when they could technically ship FCL.

If you’re unsure which option makes sense for your shipment, it’s worth checking with your forwarder or logistics provider, as the practical tipping point can vary by route, market conditions, and current capacity.

Is it better to let my supplier arrange freight or to use my own forwarder?

In most cases, shippers benefit from using their own freight forwarder, mainly for reasons of visibility and transparency.

When you work directly with a forwarder, it’s usually clearer what services are included in the quote, how costs are broken down, and who is responsible for each part of the shipment. That makes it easier to understand what you’re paying for and to spot potential gaps or add-ons before they become surprises.

When suppliers arrange freight, they typically work with logistics providers they already have relationships with. While this can be convenient, it often gives the shipper less insight into pricing and service scope, and additional charges may appear later that weren’t obvious upfront.

That said, supplier-arranged freight can make sense in some situations, especially for very small or infrequent shipments, but shippers who want more control and predictability usually prefer working with their own forwarder.

Do you need to know the seaport code for, say, the UK’s largest container port at Felixstowe? Check out this handy Seaport Code Finder. It’s GBFXT, by the way.

The post Ocean Freight Rates & Shipping Guide appeared first on Freightos.

Non classé

OpenAI’s $1 Trillion Wait Is an AI Infrastructure Supply Chain Story

Published

6 heures ago

14 juillet 2026

OpenAI’s reported consideration of a later IPO is not just a valuation debate. It exposes the capital, compute, energy, semiconductor, and data-center supply chains required to support frontier artificial intelligence.

OpenAI’s reported consideration of waiting until 2027 to complete an initial public offering is being treated primarily as a capital-markets story. The discussion has centered on timing, valuation, and whether public investors are prepared to support a company worth approximately $1 trillion.

That framing is too narrow.

OpenAI confirmed in June that it had confidentially submitted a draft S-1 registration statement to the Securities and Exchange Commission. The company said it had not decided when to proceed and indicated that some of its plans could be easier to execute while it remained private. Subsequent reporting said OpenAI’s advisers had discussed two possible paths: list sooner at a lower valuation or wait until 2027 and pursue a valuation closer to $1 trillion.

OpenAI has not publicly confirmed that it made that choice or formally delayed an offering. But for supply chain leaders, the precise IPO date is not the most important part of the story.

The larger issue is what a possible delay reveals about the physical and financial infrastructure required to support frontier AI.

The largest AI developers are no longer simply software companies. They are becoming major buyers of advanced semiconductors, cloud capacity, data centers, networking equipment, electrical power, cooling systems, and specialized engineering services. Their growth depends on an increasingly complex industrial network that extends far beyond model development.

OpenAI’s valuation is therefore inseparable from the supply chain required to support it.

AI Is Becoming an Industrial Business

Traditional enterprise software companies could scale without constructing an enormous physical asset base. Once the product was built, serving additional customers often required relatively little new infrastructure.

Frontier AI changes that model.

Training more capable models requires large clusters of accelerators, high-bandwidth memory, advanced networking, extensive datasets, and highly specialized technical talent. Operating those models for hundreds of millions of users creates a separate and continuing inference burden. Every query, generated image, video, and autonomous-agent task consumes computing capacity.

That demand must be met in real time, at scale, and with acceptable reliability.

Reuters reported that OpenAI was targeting roughly $600 billion in total compute spending through 2030, citing a person familiar with the company’s plans. OpenAI President Greg Brockman later testified that the company expected to spend approximately $50 billion on computing power in 2026. These are forward-looking estimates rather than audited results, and the actual totals could change materially.

The direction is nevertheless clear.

This is not a conventional technology procurement program. It is an industrial expansion that reaches from semiconductor fabrication and advanced packaging to server assembly, optical networking, data-center construction, power generation, transmission equipment, water management, and cooling infrastructure.

It also depends on labor markets that are already constrained. Chip designers, electricians, engineers, construction workers, grid specialists, and data-center technicians all sit somewhere in the chain.

OpenAI cannot support a trillion-dollar valuation through software adoption alone. The infrastructure behind the software must deliver enough capacity, fast enough, at a cost the business model can absorb.

Revenue Must Catch Up With Capacity

This is where the IPO discussion becomes a supply chain story.

PitchBook interprets the reported timing debate as a signal about the valuation public investors may currently be willing to support. That is PitchBook’s interpretation, not a conclusion disclosed by OpenAI. But it points directly to the company’s central operating challenge.

OpenAI must secure chips, cloud capacity, electrical power, and data-center infrastructure before all the corresponding revenue exists. It must make large commitments today based on demand that may take years to mature.

In supply chain terms, OpenAI is making long-lead-time capacity decisions against an uncertain demand forecast.

The demand for AI is real. The final revenue model is less certain.

Consumer subscriptions, enterprise contracts, application programming interfaces, advertising, commerce, and autonomous agents may all contribute. But each revenue stream has different implications for pricing, margins, utilization, and infrastructure requirements. A consumer query, an enterprise workflow, and an autonomous software agent may all use the same underlying model while producing very different economics.

Reuters reported that OpenAI generated approximately $5.7 billion in first-quarter 2026 revenue while consuming about $3.7 billion in cash, citing a report based on documents provided to shareholders. Reuters said it could not independently verify the figures.

Those reported numbers illustrate both sides of the equation. Demand is growing rapidly, but so is the cost of serving it.

OpenAI does not simply need more revenue. It needs revenue with margins and cash economics strong enough to finance the infrastructure behind the product.

That is a much harder problem.

The New Capacity Risk

Manufacturers have always understood the danger of investing ahead of demand.

Build too little capacity, and growth is constrained. Build too much, and fixed costs overwhelm margins. The problem becomes even more difficult when the assets are expensive, the lead times are long, and the underlying technology is changing quickly.

Frontier AI companies now face that same problem at extraordinary scale.

Advanced semiconductor capacity cannot be added overnight. Data centers require land, permits, transformers, construction materials, power agreements, network connectivity, and cooling systems. New power-generation and transmission projects can take years. Large infrastructure programs also depend on suppliers that are already serving other hyperscalers, utilities, governments, and industrial customers.

The risks are tightly connected.

AI developers may struggle to secure enough chips, memory, transformers, or electrical capacity. Competition can push infrastructure costs higher. More efficient models or processors can weaken the economics of assets ordered years earlier. Enterprise adoption may grow without producing the utilization or pricing needed to support existing commitments.

Supplier concentration adds another layer of exposure. Critical parts of the stack remain controlled by a relatively small group of semiconductor manufacturers, equipment suppliers, cloud platforms, networking vendors, and electrical-infrastructure providers.

None of these risks is unfamiliar to supply chain executives. What is different is the size of the commitments and the speed at which AI companies are trying to build the network.

Private Capital Is Buying OpenAI Time

Remaining private gives OpenAI more flexibility to make those investments without the same quarterly scrutiny faced by public companies. A confidential filing also allows the company to begin the regulatory process without immediately publishing a complete prospectus.

But waiting has a cost.

Every additional period spent private shifts more of the financing burden to investors, lenders, strategic partners, and infrastructure providers. OpenAI announced in March that it had raised $122 billion in committed capital at a post-money valuation of $852 billion. Reuters later reported that Bank of America had extended a $520 million credit line to the company, citing a person familiar with the transaction.

That financing is doing more than funding growth. It is buying OpenAI time.

The company can use that time to expand enterprise adoption, improve monetization, raise infrastructure utilization, and strengthen the economics it will eventually need to present to public investors.

In supply chain terms, the financing acts as a buffer against uncertainty. It is not inventory in the literal sense, but it serves a similar strategic purpose. It creates room between current commitments and the point at which the business must prove that those commitments can generate adequate returns.

The buffer also has carrying costs. Dilution, interest expense, financing complexity, and dependence on private-market valuations all increase the longer the company remains private.

OpenAI may be postponing public-market scrutiny, but it is not postponing the cost of building the system.

Anthropic Could Establish the First Benchmark

Anthropic has also confidentially submitted a draft S-1. Neither company has announced a firm IPO date, but the order in which they reach the market could matter.

If Anthropic lists first, its public filings could provide the first detailed benchmark for the economics of a frontier-model company. Investors may gain greater visibility into revenue recognition, cloud costs, gross margins, customer concentration, compute obligations, stock-based compensation, and capital efficiency.

Those disclosures would affect more than the valuations of OpenAI and Anthropic.

Cloud providers could face greater pressure to explain the profitability of their AI investments. Semiconductor suppliers could gain a clearer view of sustainable demand. Enterprise buyers could better assess whether current pricing is durable. Data-center and energy developers could begin separating committed long-term workloads from more speculative capacity reservations.

The AI sector has grown under conditions of limited financial transparency and extraordinary private-market enthusiasm. Public-market disclosure could impose a level of operating and supply chain discipline that the sector has not yet faced.

The Enterprise Lesson

The lesson for supply chain executives is straightforward: frontier AI should not be treated as an ordinary software category.

The service may be delivered through an application programming interface, but its availability, reliability, and price depend on a capital-intensive physical network. Semiconductor capacity, power availability, cloud architecture, financing conditions, and supplier concentration all influence the service the customer ultimately receives.

Strategic AI providers should therefore be evaluated like other critical suppliers.

Enterprises should examine financial durability, infrastructure partnerships, contractual protections, data portability, model-substitution options, and dependence on a single provider. They should also understand where workloads can move if capacity becomes constrained, pricing changes, or a provider alters its strategy.

Architectures that allow work to move among multiple models reduce exposure to any one company’s pricing, capacity, outages, and strategic decisions.

A multi-model strategy is not simply a technical choice. It is supply chain risk management.

A Valuation Built on Execution

A valuation approaching $1 trillion may eventually be supportable. OpenAI has broad market reach, substantial enterprise momentum, a powerful brand, and access to enormous amounts of capital.

But user growth alone will not justify it.

The company must convert adoption into durable revenue while managing one of the largest technology-infrastructure expansions ever attempted. It must secure capacity before demand is fully monetized, finance that capacity while remaining flexible, and avoid allowing infrastructure costs to overwhelm the economics of the product.

That is why the IPO debate matters.

The market is beginning to ask harder questions. Who will finance the infrastructure? How quickly will that infrastructure produce returns? Can the supporting supply chain scale without undermining the economics of the products it enables?

OpenAI may be waiting for a better market.

More fundamentally, it may be waiting for the business model to catch up with the supply chain required to deliver it.